Search papers, labs, and topics across Lattice.
2
0
3
2
Achieving state-of-the-art video matting performance with a framework trained exclusively on images reveals the untapped potential of existing trackers in fine-grained tasks.
MLLMs can now segment emerging entities with significantly improved accuracy thanks to ROSE, a retrieval-augmented framework that boosts performance by 19.2 gIoU over a Gemini-2.0 Flash baseline.
